Research - Michael Ivanitskiy

My Google Scholar

• Understanding Search in Transformers
  (primary project) mechanistic interpretability of transformer networks trained on toy spatial tasks. Latest Paper, Code
• BNBP
  Training of Hodgkin-Huxley neural nets with SGD
• Inverse Scaling
  resistance to prompt injection might be an example of a task which exhibits inverse scaling in large transformer networks
• RL dreams
  value function robustness for world-model RL by giving it weird dreams
• Loss Symmetries
  (coming soon)
• CE-learn
  Studying learning in the C. elegans nematode
• ML for medical imaging
  Segmentation of CT images using a U-Net architecture